A task processing method, apparatus and device

Abstract

The application provides a task processing method, device and equipment. The task processing method comprises the following steps: acquiring state space data of at least one terminal and at least one edge computing device when the terminal and the edge computing device interact with each other; inputting the state space data of a first time in a preset time sequence into a first processing network of a task processing model for processing to obtain first evaluation data; inputting the state space data of a second time in the preset time sequence into a second processing network of the task processing model for processing to obtain second evaluation data; the second time is greater than the first time; and updating parameters of the task processing model according to the first evaluation data and the second evaluation data to obtain an update result. The scheme of the application can utilize edge resources based on cooperation between multiple terminals and multiple edge computing devices, and can process a task through a task processing model, thereby improving the computing efficiency.

Description

Technical Field

[0001] This invention relates to the field of computer information processing technology, and in particular to a task processing method, apparatus, and device. Background Technology

[0002] With the widespread adoption of intelligent computing devices such as mobile phones and computers, and the development of artificial intelligence algorithms, the processing of various application tasks places increasingly higher demands on the real-time performance of data. For such applications, local computing may result in insufficient capacity or excessive power consumption; while cloud computing may fail to meet the application's latency requirements. As the application scales up, network bandwidth in the cloud computing architecture becomes a bottleneck, making it difficult to support large-scale real-time computing and massive data requests.

[0003] Currently, task offloading methods can be divided into binary computation offloading and partial computation offloading. Specifically, binary computation offloading refers to processing all tasks on the local device or on the server; partial offloading refers to the terminal device transferring a portion of the task to the server for processing, while processing the remaining portion locally. At least the following problems exist with task offloading:

[0004] (1) The mission arrival status does not conform to the dynamic characteristics of the actual environment;

[0005] (2) When calculating the delay, the impact of the underlying queuing system is not considered;

[0006] (3) Lack of collaboration between MEC servers;

[0007] (4) The efficiency of sampling learning is low during experience playback. Summary of the Invention

[0008] The technical problem to be solved by the present invention is how to provide a task processing method, apparatus and device that can make full use of edge resources based on the collaboration between multiple terminals and multiple edge computing devices, and process tasks through a task processing model to improve computing efficiency.

[0009] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0010] A task processing method, the method comprising:

[0011] Acquire state space data of at least one terminal interacting with at least one edge computing device during a preset time series;

[0012] The state space data at the first moment in the preset time series is input into the first processing network of the task processing model for processing to obtain the first evaluation data.

[0013] The state space data at the second moment in the preset time series is input into the second processing network of the task processing model for processing to obtain the second evaluation data; the second moment is greater than the first moment.

[0014] Based on the first evaluation data and the second evaluation data, the parameters of the task processing model are updated to obtain the update result.

[0015] Optionally, the first processing network includes a first policy network and a first evaluation network;

[0016] The process of inputting the state space data at the first moment in the preset time series into the first processing network of the task processing model for processing to obtain the first evaluation data includes: inputting the state space data at the first moment into the first policy network for processing to obtain the action state data at the first moment.

[0017] The state space data and action state data at the first moment are input into the first evaluation network for processing to obtain the first evaluation data.

[0018] Optionally, after inputting the state space data at the first moment into the first policy network for processing to obtain the action state data at the first moment, the method further includes:

[0019] Based on the action status data at the first moment, determine the latency and energy consumption data when at least one terminal interacts with at least one edge computing device.

[0020] Optionally, the second processing network includes a second policy network and a second evaluation network;

[0021] Specifically, the state space data at the second moment in the preset time series is input into the second processing network of the task processing model for processing to obtain the second evaluation data, including:

[0022] The state space data at the second time step is input into the second policy network for processing to obtain the action state data at the second time step.

[0023] The state space data and action state data at the second time point are input into the second evaluation network for processing to obtain the second evaluation data.

[0024] Optionally, based on the first evaluation data and the second evaluation data, the parameters of the task processing model are updated to obtain an update result, including:

[0025] Based on the latency and energy consumption data and the first evaluation data, the parameters of the first policy network of the task processing model are updated to obtain the first update result;

[0026] By minimizing the loss function, the parameters of the first evaluation network of the task processing model are updated to obtain the second update result;

[0027] Based on the first update result, the parameters of the second policy network of the task processing model are updated to obtain the third update result;

[0028] Based on the second update result, the parameters of the second evaluation network of the task processing model are updated to obtain the fourth update result.

[0029] Optionally, based on the first update result, the parameters of the second policy network of the task processing model are updated to obtain a third update result, including:

[0030] According to θ π′ =τθ π +(1-τ)θ π′ The parameters of the second policy network of the task processing model are updated to obtain the third update result;

[0031] Where, θ π′ θ represents the parameters of the second policy network in the task processing model. π This is the first update result, and τ is a preset variable.

[0032] Optionally, based on the second update result, the parameters of the second evaluation network of the task processing model are updated to obtain a fourth update result, including:

[0033] According to θ Q′ =τθ Q +(1-τ)θ Q′ The parameters of the second evaluation network of the task processing model are updated to obtain the fourth update result;

[0034] Where, θ Q′ θ represents the parameters of the second evaluation network for the task processing model. Q This is the second update result, where τ is a preset variable.

[0035] The present invention also provides a task processing apparatus, the apparatus comprising:

[0036] The acquisition module is used to acquire state space data of at least one terminal interacting with at least one edge computing device during a preset time series.

[0037] The processing module is used to input the state space data at the first moment in the preset time series into the first processing network of the task processing model for processing to obtain the first evaluation data.

[0038] The state space data at the second moment in the preset time series is input into the second processing network of the task processing model for processing to obtain the second evaluation data; the second moment is greater than the first moment.

[0039] Based on the first evaluation data and the second evaluation data, the parameters of the task processing model are updated to obtain the update result.

[0040] The present invention also provides a computing device, comprising: a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method described above.

[0041] The present invention also provides a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described above.

[0042] The above-described solution of the present invention has at least the following beneficial effects:

[0043] The above-described solution of the present invention obtains state space data of at least one terminal interacting with at least one edge computing device during task processing within a preset time series; inputs the state space data at a first moment in the preset time series into a first processing network of a task processing model for processing to obtain first evaluation data; inputs the state space data at a second moment in the preset time series into a second processing network of the task processing model for processing to obtain second evaluation data; the second moment is greater than the first moment; and updates the parameters of the task processing model based on the first evaluation data and the second evaluation data to obtain an update result. This approach enables full utilization of edge resources based on collaboration between multiple terminals and multiple edge computing devices, and improves computational efficiency by processing tasks through a task processing model. Attached Figure Description

[0044] Figure 1 This is a flowchart illustrating the task processing method provided in an embodiment of the present invention;

[0045] Figure 2 This is an architecture diagram of the task processing system according to an embodiment of the present invention;

[0046] Figure 3 This is a schematic diagram of the processing process of the task processing model in an embodiment of the present invention;

[0047] Figure 4 This is a schematic diagram of the module block of the task processing device according to an embodiment of the present invention. Detailed Implementation

[0048] Exemplary embodiments of the present application will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0049] like Figure 1 As shown, an embodiment of the present invention provides a task processing method, the method comprising:

[0050] Step 11: Obtain state space data of at least one terminal interacting with at least one edge computing device during a preset time series.

[0051] Step 12: Based on the state space data at the first moment in the preset time series, input the data into the first processing network of the task processing model for processing to obtain the first evaluation data;

[0052] Step 13: Based on the state space data at the second moment in the preset time series, input it into the second processing network of the task processing model for processing to obtain the second evaluation data; the second moment is greater than the first moment;

[0053] Step 14: Update the parameters of the task processing model based on the first evaluation data and the second evaluation data to obtain the update result.

[0054] In this embodiment of the present invention, the task processing method can fully utilize edge resources based on the collaboration between multiple terminals and multiple edge computing devices, and process tasks through a task processing model, thereby improving computing efficiency.

[0055] It should be noted that the terminal includes, but is not limited to: mobile phones, computers, tablets, smartwatches, etc.; the edge computing device includes: at least one base station, and an edge server, i.e., an MEC (Mobile Edge Computing) server, set up next to at least one of the base stations; and task processing includes, but is not limited to, task unloading processing.

[0056] It should be noted that in the current MEC architecture, network deployment is based on Orthogonal Frequency Division Multiple Access (OFDMA). Since the wireless communication channels within the system are orthogonally separated, mutual interference between channels can be effectively mitigated. Therefore, the transmission rate between user i and base station m in time slot t can be calculated as follows:

[0057]

[0058] Among them, Bi (t) represents the bandwidth allocated to user i in time slot t, P i (t) represents the transmission power of device i in time slot t, g i,m (t) represents the channel gain between device i and base station m in time slot t, σ represents the noise power of data transmission, and x i,m (t) is A i (t) Whether to unload to MEC server m, A i (t) represents the application generated by terminal i in time slot t, constrained to choose only one MEC server. If it is offloaded to MEC server M... m Then set x i,m (t) = 1, otherwise x i,m (t) = 0, For all with A i The subtasks of (t) arrive simultaneously and select the same MEC server from other devices.

[0059] It is worth noting that when processing tasks, the terminal can choose to send part of the task to the edge computing device for processing. This allows the edge computing device to help the terminal to process tasks collaboratively, thereby improving task processing efficiency.

[0060] The terminal can send some tasks to an attached edge computing device for processing, or it can send some tasks to other edge computing devices for processing. If the terminal is located within the effective service radius of the edge computing device, then the edge computing device is an attached edge computing device of the terminal. When the terminal sends some tasks to an attached edge computing device for processing, the terminal can directly send some tasks to the attached edge computing device for processing. When the terminal sends some tasks to other target edge computing devices for processing, the terminal first sends some tasks to the attached edge computing device, and then the attached edge computing device sends some tasks to the target edge computing device for processing.

[0061] Specifically, such as Figure 2 As shown, in time slots t∈{1,2,3…T}, the time slot length is defined as Δ. Since the task processing model better reflects the dynamic nature of the environment, it is assumed that the arrival of applications follows a Poisson distribution, and λ... i (t) is defined as the task arrival rate of terminal i in time slot t, A i (t) represents the application generated by terminal i in time slot t, assuming that application A i (t) can be divided into multiple independent subtasks, which can run locally on the terminal or be offloaded to the MEC server for execution; the application set is: A = {A i (1),A i(2)....,A i (t)…,A i (t)}. The application's attributes are represented by the tuple (W i (t),I i (t)) constitutes. W i (t) represents the load newly arriving in the application at the start of time slot t, I i (t) represents W i (t) Input data size. If no application arrives at the start of time slot t, then W i (t)=I i (t) = 0. Let ξ be the CPU (Central Processing Unit) processing density cpb, representing the number of clock cycles (cycles / bit) required for the processor to execute each bit. Then W i (t) and I i The relationship between (t) is:

[0062] W i (t)=ξ×I i (t)

[0063] Application A i (t) can be roughly divided into n independent tasks. And calculate the percentage of data volume for each part of the task. and satisfy definition A represents i (t) represents the set of percentages that can be selected for uninstallation. Define p i (t) is A i (t) Actual selected unloading rate, Where r = 0 or 1, and satisfies 0 < p i (t)≤1 indicates that the task will definitely be unloaded to the MEC server.

[0064] Assume the effective service radius of the edge computing device is R, and the Euclidean distance between the terminal device and the edge computing device is Dis(U) i M m ). Using sets Indicates terminal device U i You can directly request services from MEC servers. If the selected MEC server is in this set, the terminal device will allocate 1-p based on the offload ratio. i (t) The portion processed locally is placed into the local computation queue. p i Part of (t) is placed in the local transmission queue. And transmits the computation queue to the target MEC server via a wireless link. If the selected MEC server is not in the set, then define the cooperative scheduling factor. M m' Indicates the target MEC server. This indicates the MEC server closest to the end user, when When, it indicates that the terminal device U i The subtask needs to be passed through the nearest To forward, first, p i The part of (t) is placed into transmission queue Then, the transmission queue forwards it to M. m' computation queue

[0065] In an optional embodiment of the present invention, the first processing network includes a first policy network and a first evaluation network; step 12 may include:

[0066] Step 121: Input the state space data at the first moment into the first policy network for processing to obtain the action state data at the first moment;

[0067] Step 122: Input the state space data and the action state data at the first moment into the first evaluation network for processing to obtain the first evaluation data.

[0068] In a specific implementation, after step 121, the following may also be included:

[0069] Step 1211: Based on the action state data at the first moment, determine the latency and energy consumption data when at least one terminal interacts with at least one edge computing device.

[0070] In this embodiment, state space data at a certain moment is input into a first policy network for processing to obtain action state data at the first moment. The action state data and the state space data are then input into a first evaluation network for processing to obtain first evaluation data. This allows the output of the first evaluation data, thereby achieving the evaluation of the action state data.

[0071] It should be noted that, based on the action state data at the first moment, the latency and energy consumption data when at least one terminal interacts with at least one edge computing device are calculated. According to the latency and energy consumption data, the task interaction process between at least one terminal and at least one edge computing device at the next moment is carried out. In this way, the state space data at the next moment can be obtained. The state space data, action state data and experimental energy consumption data at each moment can be used to generate an empirical sample set, which is convenient for providing input data for the task processing model.

[0072] Specifically, when terminals and edge computing devices interact with each other, it is necessary to determine the impact of the local computing queue, local transmission queue, edge computing queue, and edge transmission queue on task processing.

[0073] The process of determining the local computing queue includes:

[0074] Each application contains a different amount of data. The processing of each task on the local processor is a single service process, and these processes are independent of each other. Assuming the local processor uses DVFS (Dynamic Voltage and Frequency Scaling) technology, its computing power is determined by its clock frequency. This is measured by the number of CPU revolutions per unit time. The latency of a subtask being computed locally is defined as the time elapsed from its creation to complete processing. The latency required for a subtask to be computed locally is determined by the waiting latency in the queue. Service delay composition:

[0075]

[0076]

[0077] in, To calculate the time required for queue processing locally, This represents the number of time slots the subtask needs to wait in the local queuing system. Let t be the time when all subtasks placed in the local computation queue before time slot t are processed; for the computation (or transmission) queue, if the processing (or transmission) of a subtask is completed in one time slot, then the next subtask in the queue will be processed (or transmitted) at the beginning of the next time slot. equal Subtract the arrival time of the subtask and add 1. The following formula represents the time slot in which the subtask will be processed in the local computation queue. The internal organs have been processed.

[0078]

[0079] in, This represents the floor function. This indicates the number of time slots required for the subtask to be processed in the local computation queue, at time 0. Initialization occurs during which the subtasks do not need to wait, therefore Therefore

[0080] The energy consumed by the terminal device when performing local computing tasks is:

[0081]

[0082] in, This represents the energy consumed per CPU cycle locally, typically taken as a value.

[0083] The process of determining the local transmission queue includes:

[0084] For most applications, the output data size is typically much smaller than the input, and the downlink data rate is much higher than the uplink data rate. Therefore, the latency of transmitting the result back to the user end is omitted in the calculation, and the downlink time is negligible. Thus, the time required to send the calculation or transmission queue from the local transmission queue to the MEC server consists of the local transmission queue's waiting latency and the transmission latency.

[0085]

[0086]

[0087] in, A represents i (t) The number of time slots that the subtask to be transmitted to MEC needs to wait in the local transmission queue. This indicates the time slot in which all subtasks placed in the transmission queue before time slot t have been processed.

[0088] The following formula indicates that the subtask will be in the time slot in the local transmission queue. The internal organs have been processed.

[0089]

[0090] in, This indicates the number of time slots required for a subtask to be processed in the local transmission queue, at time 0. Initialization, at this time Therefore

[0091] The energy consumed by the terminal device when transmitting tasks locally is:

[0092]

[0093] Among them, P trans This indicates the device's transmission power.

[0094] The process of determining the edge computing queue includes:

[0095] Assuming the time when the unloaded subtask arrives at the edge computing queue is t^, the time required for it to be in the MEC computing queue consists of the task processing time and the waiting time in the computing queue, that is:

[0096]

[0097] in, This represents the number of time slots the unloaded subtask needs to wait in the MEC compute queue, and it is calculated as follows:

[0098]

[0099] in, This represents the length of the queue at the start of time slot t^, and its update formula is as follows:

[0100]

[0101] in, It is the queue length of MEC server m starting in time slot t^-1. This represents the sum of tasks from all terminal devices arriving at the MEC server m at time t^-1. This indicates the amount of data that the MEC server m can process in a single time slot.

[0102] Since applications from different terminal devices may arrive at the same server simultaneously, the MEC server will add them to the computation queue in a random order. Therefore, a variable is defined. To represent device i in the MEC compute queue * The processing order of the tasks in the two parts: device i and device i.

[0103]

[0104] Therefore Indicates all and A i The subtasks of (t) arrive at the same MEC server computation queue simultaneously and are computed before A. i (t) The sum of the quantities of other subtasks of the subtask.

[0105] The energy consumed by the terminal device when the MEC server is performing computational tasks is:

[0106]

[0107] The process of determining the edge transmission queue includes:

[0108] At the beginning of time slot t, when the new application A... iWhen task (t) arrives, it can be processed locally, offloaded to an adjacent edge computing device, or further forwarded to other nearby edge computing devices via a zero-latency wired link in the network. Tasks that need to be transferred to nearby edge computing devices are placed in the transmission queue of the MEC server configured on the directly connectable edge computing device. Assume that the time when the offloaded subtask arrives at the edge transmission queue is t. * Edge computing devices transmit information via wired connections at a transmission rate of Ω. The time required for an uninstalled application to be forwarded from a currently directly connectable edge computing device to the target MEC server is expressed as:

[0109]

[0110]

[0111] in, This indicates the number of time slots the unloaded subtask needs to wait in the MEC server's transmission queue. Indicates time slot t * The time slot when all subtasks previously placed in the MEC server's transmission queue have been processed.

[0112] The following formula indicates that the subtask will be in the time slot in the MEC server's transmission queue. The internal organs have been processed.

[0113]

[0114] in, This indicates the number of time slots required for the unloaded subtask to be processed in the MEC server's transport queue, at time 0. Initialization is performed at this time. Therefore

[0115] The energy consumed by the terminal device when transmitting tasks to the MEC server is:

[0116]

[0117] Among them, P idle This indicates the power of the device when it is idle.

[0118] The weighted sum of energy consumption and latency required for a terminal device to execute a task reflects the overhead incurred in offloading the task. The time overhead for offloading an application between terminal i in time slot t and the attached edge computing device is:

[0119]

[0120] The time overhead for unloading the application between the attached edge computing device and the target edge computing device at terminal i in time slot t is:

[0121]

[0122] The energy consumption cost for offloading the application between terminal i in time slot t and the attached edge computing device is:

[0123]

[0124] The energy consumption cost for offloading applications between the attached edge computing device and the target edge computing device at terminal i in time slot t is:

[0125]

[0126] When the target MEC server selected by the terminal device is not within its connection range, it needs to be forwarded through the edge computing device it is connected to. At this time, the total overhead increases the time and energy consumption of MEC transmission compared to when there is no cooperation.

[0127] In a multi-terminal device, multi-MEC server architecture, considering queue waiting time constraints, task scheduling constraints, and MEC server computing resource constraints, the optimization objective is to minimize the total overhead across all time slots for all user devices within the system. Therefore, this paper establishes the following model:

[0128]

[0129] st

[0130]

[0131]

[0132] C3:μ+v=1

[0133]

[0134] Where μ is the time weighting factor, with the unit being joules per second, used to convert the unit of time consumption into the unit of energy consumption; v is the energy weighting factor, which is dimensionless; C1 indicates that the task offloading ratio cannot exceed 1; C2 indicates that each task can only be processed on one MEC server; C3 indicates that the sum of the time ratio μ and the energy ratio v must be 1; and C4 indicates the constraint on the waiting time of tasks under each queue.

[0135] In another optional embodiment of the present invention, the second processing network includes a second policy network and a second evaluation network; step 13 may include:

[0136] Step 131: Input the state space data at the second time moment into the second policy network for processing to obtain the action state data at the second time moment;

[0137] Step 132: Input the state space data and the action state data at the second time moment into the second evaluation network for processing to obtain the second evaluation data.

[0138] In this embodiment, the state space data at the second time moment is input into the second policy network for processing to obtain the action state data at the second time moment. The action state data and the state space data are then input into the second evaluation network for processing to obtain the second evaluation data. This allows the output of the second evaluation data, thereby achieving the evaluation of the second action state data.

[0139] In another optional embodiment of the present invention, step 14 may include:

[0140] Step 141: Based on the latency and energy consumption data and the first evaluation data, update the parameters of the first policy network of the task processing model to obtain the first update result;

[0141] Step 142: Update the parameters of the first evaluation network of the task processing model by minimizing the loss function to obtain the second update result;

[0142] Step 143: Based on the first update result, update the parameters of the second policy network of the task processing model to obtain the third update result;

[0143] Step 144: Based on the second update result, update the parameters of the second evaluation network of the task processing model to obtain the fourth update result.

[0144] In a specific implementation, step 143 may include:

[0145] Step 1431, according to θ π′ =τθ π +(1-τ)θ π′ The parameters of the second policy network of the task processing model are updated to obtain the third update result;

[0146] Where, θ π′ θ represents the parameters of the second policy network in the task processing model. π This is the first update result, and τ is a preset variable.

[0147] Furthermore, step 144 may include:

[0148] Step 1441, according to θ Q′ =τθ Q +(1-τ)θ Q′The parameters of the second evaluation network of the task processing model are updated to obtain the fourth update result;

[0149] Where, θ Q′ θ represents the parameters of the second evaluation network for the task processing model. Q This is the second update result, where τ is a preset variable.

[0150] In this embodiment, by updating the parameters of the task processing model, the latency and energy consumption during task processing can be optimized, thereby improving the system's operating efficiency. For example... Figure 3 As shown, in an optional specific embodiment of the present invention, the specific implementation process of the task processing method includes:

[0151] During the task interaction between the terminal and the edge computing device, all task processing models consist of four networks: the first policy network Actor, the first evaluation network Critic, the second policy network TargetActor, and the second evaluation network TargetCritic. The detailed structure of the i-th task processing model is shown as an example.

[0152] During execution, each task processing model operates independently, and each task processing model has a first policy network. The observed values ​​s of the current interaction environment between the terminal and the edge computing device i (t) is used as input, based on the output action state data a of the first policy network. i After (t), the time delay energy consumption data r will be obtained. i (t), and at the same time the state changes, the second policy network of the task processing model obtains the observation s at the next time step. i (t+1), a single set of experiences (s) obtained from the interaction between each task processing model and the terminal with the edge computing device's environment. i (t),a i (t),r i (t,s i (t+1)) is stored in the local experience cache pool. The above process is repeated continuously, and the priority, sampling probability and importance sampling weight of each single experience are calculated to form an experience sample set.

[0153] The importance of an experience sample is measured by the specific difference between the predicted value of the current task processing model and the true value function. A larger difference indicates that the prediction of the current task processing model needs more correction, thus indicating that the experience sample is more important. Therefore, TD-error (temporal difference) is selected as the criterion for judging the importance of experience samples. It is believed that the larger the absolute value of the error, the greater its contribution to gradient learning. Experience samples with larger TD-errors in the experience pool should be sampled first to achieve priority playback of important experience data. The task processing model must consider not only its own state and behavior but also the influence of the actions of other task processing models. Therefore, single sets of experiences at the same time are taken from the local experience pools of other task processing models and concatenated to obtain experience (s(t), a(t), r(t), s(t+1)). The TD-error of the experiences in the local experience pool of task processing model i is calculated as follows:

[0154]

[0155] Among them, y h The true value function is calculated as follows:

[0156]

[0157] Where Q represents the quality function, used to evaluate the quality of action state data, and is output by the first evaluation network Critic. (s(t), a(t)) is input into the first evaluation network Critic to obtain... Inputting s(t+1) into the second policy network TargetActor of task processing model i, the output a(t+1) is concatenated with s(t+1) to obtain (s(t+1),a(t+1)), and then input into the second evaluation network TargetCritic to obtain...

[0158] To prevent network overfitting and ensure that even experiences with TD-error = 0 can be extracted, a probability-based training sample sampling technique is used. Experience is extracted probabilistically, with a sampling probability of O0. h The calculation method is as follows:

[0159]

[0160] o h =|δ h |+ε

[0161] Where H represents the total number of samples in the current local experience pool, ε is a small positive constant, and α represents the importance of the ranking value. When α = 0, it means that all samples contribute equally to the sampling, i.e., uniform sampling. When α is infinitely large, only 0...h Only the largest sample will be sampled.

[0162] In PER (Priority Experience Retrieval), the absolute value δ of TD-error is used to prioritize experience samples, serving as an indicator of whether an experience sample is worth sampling. This allows sampling of stored experience samples based on the value of δ. However, this can lead to frequent extraction and replay of experiences with high δ values, potentially reducing the diversity of training samples and causing the training process of the task processing model to oscillate or even diverge. Therefore, importance sampling weights are used to correct this. The following formula represents the importance sampling weights based on TD-error priority:

[0163] w h = (H·O) h ) -β

[0164] Among them, w h Let w be the importance sampling weight corresponding to the h-th experience group, and β be a parameter representing the degree of correction. When β = 0, it means that the importance sampling weight strategy is not used. When β = 1, the weight will eliminate the bias caused by the PER sampling probability, but it may also slow down the convergence speed of the model. Therefore, adjusting β∈(0,1) can make the sampling effective and the model can converge effectively. To improve the stability during convergence, w is... h = (H·O) h ) -β After normalization, we get the following formula:

[0165]

[0166] If historical data were directly selected based on empirical probability sorting during each sampling, the computational load would be significantly increased. To further improve sampling speed, a Sumtree (binary tree) data structure is used to store empirical priority values. δ h In a Sumtree, non-leaf nodes store the values ​​of each leaf node. The root node is the sum of the storage priorities of all leaf nodes. When storing priorities, starting from a leaf node, each time data is stored, the corresponding sample pool data and the root node are updated. During data sampling, the root node's priority is divided by the number of samples K to be extracted, dividing the priorities into K intervals from 0 to the sum of priorities. Then, a number is randomly selected from each interval, and the search proceeds downwards according to a certain pattern. Each time a leaf node is sampled, its priority and the corresponding sample pool data are returned.

[0167] After all task processing models and terminals have finished interacting with the edge computing devices, the task processing models are trained using the experience sample data in the experience pool. According to the sumtree structure, K sets of samples are extracted from the local cache pool of task processing model i, and single sets of experience at the same time are taken from the local experience pools of the other task processing models. These are then concatenated to obtain K sets of experience (s(t), a(t), r(t), s(t+1)). The parameters of the first policy network Actor are then updated using gradient descent. The gradient calculation formula is:

[0168]

[0169] The input to the first evaluation network Critic for all task processing models is the same: the set of states and actions of all K groups of task processing models. Assume the j-th extracted experience is represented as (s... j (t),a j (t)), where The value network Q(s,a;θ) is updated by minimizing the loss value. Q The formula for calculating the loss is:

[0170]

[0171] Among them, w j The importance sampling weights are assigned to the j-th group of experiences extracted.

[0172] The parameters of the second policy network TargetActor and the second evaluation network TargetCritic of task processing model i are respectively All updates are performed using a soft update method, that is:

[0173]

[0174] Furthermore, the process of unloading the task includes:

[0175] like Figure 3 As shown, assuming that applications on all terminal devices arrive randomly and follow a Poisson distribution, for an application arriving on a terminal device in a certain time slot, it can be divided and offloaded according to a certain proportion. The offloaded data is sent to the target MEC server via wireless transmission. An offloading decision is made based on the target MEC server and the target proportion.

[0176] In uncertain and stochastic environments, most decision problems can be modeled using Markov Decision Processes (MDPs). An MDP consists of three types of data.<S,A,R> Let S be the state space data, which is a set of environmental information; A is the action state data, which is a set of actions performed by the task processing model based on the environmental information; and R is the time delay and energy consumption data, which represents the immediate reward value obtained after performing an action in state s(t).

[0177] For state-space data S, since the goal of reinforcement learning is to continuously learn policies from historical information, defining a comprehensive state is crucial for decision-making efficiency. Treating each terminal device as a task processing model, and considering factors such as the amount of data applied, network conditions, and computing resources, the state of the i-th task processing model in time slot t is defined as follows:

[0178]

[0179] Among them, W i (t) represents the workload of terminal device i in time slot t. Represents local computing power, {r i,1 (t),r i,2 (t),…,e i,m (t)} represents the uplink transmission rate between terminal device i and all MEC servers. This indicates the amount of data in the computation queues of all MEC servers that are pending processing.

[0180] For action state data A, each action of the task processing model includes a set of unloading behaviors. When each terminal receives state s... i When (t), it will target the MEC server and the target proportion p. i (t) makes a decision, defining the action space as a i (t)=(M i (t),p i (t)).

[0181] For instantaneous experimental energy consumption data R, in each time slot t, after the task processing model performs action a(t) in a certain state s(t), it will receive a reward R(s(t), a(t)). In practice, the reward function should be positively correlated with the objective function. The reward mechanism is negatively correlated with the total task processing cost; that is, the larger the reward, the lower the total processing cost. The instantaneous reward obtained by the task processing model in time slot t is R. i (t)=-(αT i (t)+βE i (t)). The total reward obtained by all task processing models in time slot t can be expressed as:

[0182] For each task processing model, the input to the first policy network Actor is its state, denoted as s. i (t), then through two fully connected layers, and finally through a softmax layer in the output layer, each output value represents the probability of the corresponding action. The input to the first evaluation network Critic is the state (s1(t),…,s2(t),…,s) of all task processing models. I (t) and actions (a1(t),...,a2(t),...,a I (t), consisting of two fully connected hidden layers and an output layer with one node, outputs the Q value. The network structures of the second policy network TargetActor and the second evaluation network TargetCritic are the same as those of the first policy network Actor and the first evaluation network Critic, respectively, and their network parameters are updated using a soft update method.

[0183] In the above embodiments of the present invention, the systems coordinate with each other through mutual communication, solve problems in parallel, and obtain the optimal strategy through iterative learning of historical experience, thereby improving the system's operating efficiency. The task processing method introduces priority experience replay and importance sampling into the MADDPG algorithm to improve sampling efficiency.

[0184] Since the PER-MADDPG algorithm is based on DDPG, and DDPG depends on Therefore, a continuous action space is required. However, in this paper, the action space consists of the target MEC server and the offloading ratio, both of which are discrete values. To better suit this scenario, the deterministic strategy of the Actor network in MADDPG was removed, and the output layer was changed to a network that can output the probabilities of all actions, making the algorithm usable in a discrete action space.

[0185] like Figure 4 As shown, embodiments of the present invention also provide a task processing device 40, the device 40 comprising:

[0186] The acquisition module 41 is used to acquire state space data of at least one terminal interacting with at least one edge computing device during a preset time series.

[0187] Processing module 42 is used to input the state space data at the first moment in the preset time series into the first processing network of the task processing model for processing to obtain the first evaluation data.

[0188] The state space data at the second moment in the preset time series is input into the second processing network of the task processing model for processing to obtain the second evaluation data; the second moment is greater than the first moment.

[0189] Based on the first evaluation data and the second evaluation data, the parameters of the task processing model are updated to obtain the update result.

[0190] Optionally, the first processing network includes a first policy network and a first evaluation network;

[0191] The process of inputting the state space data at the first moment in the preset time series into the first processing network of the task processing model for processing to obtain the first evaluation data includes: inputting the state space data at the first moment into the first policy network for processing to obtain the action state data at the first moment.

[0192] The state space data and action state data at the first moment are input into the first evaluation network for processing to obtain the first evaluation data.

[0193] Optionally, after inputting the state space data at the first moment into the first policy network for processing to obtain the action state data at the first moment, the method further includes:

[0194] Based on the action status data at the first moment, determine the latency and energy consumption data when at least one terminal interacts with at least one edge computing device.

[0195] Optionally, the second processing network includes a second policy network and a second evaluation network;

[0196] Specifically, the state space data at the second moment in the preset time series is input into the second processing network of the task processing model for processing to obtain the second evaluation data, including:

[0197] The state space data at the second time step is input into the second policy network for processing to obtain the action state data at the second time step.

[0198] The state space data and action state data at the second time point are input into the second evaluation network for processing to obtain the second evaluation data.

[0199] Optionally, based on the first evaluation data and the second evaluation data, the parameters of the task processing model are updated to obtain an update result, including:

[0200] Based on the latency and energy consumption data and the first evaluation data, the parameters of the first policy network of the task processing model are updated to obtain the first update result;

[0201] By minimizing the loss function, the parameters of the first evaluation network of the task processing model are updated to obtain the second update result;

[0202] Based on the first update result, the parameters of the second policy network of the task processing model are updated to obtain the third update result;

[0203] Based on the second update result, the parameters of the second evaluation network of the task processing model are updated to obtain the fourth update result.

[0204] Optionally, based on the first update result, the parameters of the second policy network of the task processing model are updated to obtain a third update result, including:

[0205] According to θ π′ =τθ π +(1-τ)θ π′ The parameters of the second policy network of the task processing model are updated to obtain the third update result;

[0206] Where, θ π′ θ represents the parameters of the second policy network in the task processing model. π This is the first update result, and τ is a preset variable.

[0207] Optionally, based on the second update result, the parameters of the second evaluation network of the task processing model are updated to obtain a fourth update result, including:

[0208] According to θ Q′ =τθ Q +(1-τθ Q′ The parameters of the second evaluation network of the task processing model are updated to obtain the fourth update result;

[0209] Where, θ Q′ θ represents the parameters of the second evaluation network for the task processing model. Q This is the second update result, where τ is a preset variable.

[0210] It should be noted that this device is the same as the method described above. All implementations in the above method embodiments are applicable to the embodiments of this device and can achieve the same technical effect.

[0211] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0212] Embodiments of the present invention also provide a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0213] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0214] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0215] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0216] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0217] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0218] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0219] Furthermore, it should be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent solutions of the present invention. Moreover, the steps performing the above-described series of processes can naturally be executed in the order described, but are not necessarily required to be executed in chronological order; some steps can be executed in parallel or independently of each other. Those skilled in the art will understand that all or any step or component of the method and apparatus of the present invention can be implemented in any computing device (including processors, storage media, etc.) or network of computing devices, in hardware, firmware, software, or a combination thereof. This is something that those skilled in the art can achieve by using their basic programming skills after reading the description of the present invention.

[0220] Therefore, the object of the present invention can also be achieved by running a program or a set of programs on any computing device. The computing device can be a known general-purpose device. Therefore, the object of the present invention can also be achieved simply by providing a program product containing program code implementing the method or apparatus. That is, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any known storage medium or any storage medium developed in the future. It should also be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent to the present invention. Furthermore, the steps performing the above series of processes can naturally be performed in the order described, but are not necessarily required to be performed in chronological order. Some steps can be performed in parallel or independently of each other.

[0221] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A task processing method, characterized in that, The method includes: Obtain state space data of at least one terminal interacting with at least one edge computing device during a preset time series; the edge computing device includes: a base station and an edge server located next to the base station; the task includes: an application generated by terminal i in time slot t. Multiple independent subtasks ; The state space data at the first moment in the preset time series is input into the first processing network of the task processing model for processing to obtain the first evaluation data. The state space data at the second moment in the preset time series is input into the second processing network of the task processing model for processing to obtain the second evaluation data; the second moment is greater than the first moment. Based on the first evaluation data and the second evaluation data, the parameters of the task processing model are updated to obtain an update result; wherein, the first processing network includes a first policy network and a first evaluation network; Specifically, the state space data at the first moment in the preset time series is input into the first processing network of the task processing model for processing to obtain the first evaluation data, including: The state space data at the first moment is input into the first policy network for processing to obtain the action state data at the first moment; The state space data and action state data at the first moment are input into the first evaluation network for processing to obtain the first evaluation data; The process includes, after inputting the state space data at the first moment into the first policy network for processing to obtain the action state data at the first moment, the method further includes: Based on the action status data at the first moment, determine the latency and energy consumption data when at least one terminal interacts with at least one edge computing device. When terminals and edge computing devices interact with each other, determine the impact of local computing queues, local transmission queues, edge computing queues, and edge transmission queues on task processing; The process of determining the local computing queue includes: The latency required for locally computed subtasks is determined by the waiting latency in the queue. Service delay composition: in, To calculate the time required for queue processing locally, This represents the number of time slots the subtask needs to wait in the local queuing system. ξ represents the time when all subtasks placed in the local computation queue before time slot t have been processed; ξ is the central processing unit (CPU) processing density, representing the number of clock cycles required for each bit of the processor to execute. express Input data size, This represents the amount of load newly arriving in the application at the start of time slot t. This represents the number of CPU revolutions per unit of time. for The actual selected uninstallation rate; For computation queues or transmission queues, if the processing or transmission of a subtask is completed within a time slot, then the next subtask in the queue will be processed or transmitted at the beginning of the next time slot. equal Subtract the arrival time of the subtask and add 1. The following formula indicates when the subtask will be in the local computation queue. The internal organs have been processed. in, This represents the floor function. This indicates the number of time slots required for the subtask to be processed in the local computation queue, at time 0. Initialization occurs during which the subtasks do not need to wait, therefore Therefore ; The energy consumed by the terminal device when performing local computing tasks is: ; in, This represents the energy consumed per CPU cycle locally, with a value of [value missing]. ; The process of determining the local transmission queue includes: The time required to send data from the local transmission queue to the edge server's computation or transmission queue consists of the local transmission queue's waiting latency and the transmission latency: in, express The number of time slots that a subtask destined for the edge server needs to wait in the local transmission queue. This indicates the time slot in which all subtasks placed in the transmission queue before time slot t have been processed; Δ is the time slot length. The transmission rate between user i and base station m in time slot t; The subtask will be in the local transfer queue The internal organs have been processed. in, This indicates the number of time slots required for a subtask to be processed in the local transmission queue, at time 0. Initialization, at this time Therefore ; The energy consumed by the terminal device when transmitting tasks locally is: ;in, Indicates the terminal transmission power; The process of determining the edge computing queue includes: Assume the time when the unloaded subtask arrives at the edge computing queue is The time required for a task to be processed in the edge server's computing queue consists of the task processing time and the waiting time in the computing queue, i.e.: in, This represents the number of time slots that the unloaded subtask needs to wait in the edge computing queue, and it is calculated as follows: in, Indicates in time slot The initial length of the queue is updated using the following formula: in, Is edge server m in time slot Initial queue length, express The sum of tasks from all terminal devices arriving at MEC server m at any given time. This indicates the amount of data that edge server m can process in a single time slot; Define a variable To represent devices in the edge computing queue and equipment The processing order of the two tasks; Therefore Indicates all and The subtasks arrive at the same edge server computation queue simultaneously and are computed in a priority order. The sum of the quantities of the other subtasks of the subtask; The energy consumed by the terminal device when the edge server is computing tasks is: The process of determining the edge transmission queue includes: In the time slot t At the beginning, when the new application Upon arrival, it is either processed locally or offloaded to an attached edge computing device, or further forwarded to other nearby edge computing devices via a zero-latency wired link in the network; Tasks that need to be transferred to nearby edge computing devices are placed in the transmission queue of the edge server configured on the directly connectable edge computing device. Assuming the time when the unloaded subtask arrives at the edge transmission queue is... ; Edge computing devices transmit information via wired connections at a transmission rate of [missing information]. The time required for the uninstalled application to be forwarded from the currently directly connectable edge computing device to the target edge server is expressed as: in, This indicates the number of time slots the unloaded subtask needs to wait in the edge server's transmission queue. Indicates in time slot The time slot in which all subtasks previously placed in the MEC server's transmission queue have been processed; The subtask will be in the edge server transmission queue The internal organs have been processed. in, This indicates the number of time slots required for the unloaded subtask to be processed in the edge server's transmission queue, at time 0. Initialization is performed at this time. Therefore ; The energy consumed by the terminal device when transmitting tasks to the edge server is: ;in, This indicates the power of the device when it is idle.

2. The task processing method according to claim 1, characterized in that, The second processing network includes a second policy network and a second evaluation network; Specifically, the state space data at the second moment in the preset time series is input into the second processing network of the task processing model for processing to obtain the second evaluation data, including: The state space data at the second time step is input into the second policy network for processing to obtain the action state data at the second time step. The state space data and action state data at the second time point are input into the second evaluation network for processing to obtain the second evaluation data.

3. The task processing method according to claim 1, characterized in that, Based on the first evaluation data and the second evaluation data, the parameters of the task processing model are updated to obtain the update results, including: Based on the latency and energy consumption data and the first evaluation data, the parameters of the first policy network of the task processing model are updated to obtain the first update result; By minimizing the loss function, the parameters of the first evaluation network of the task processing model are updated to obtain the second update result; Based on the first update result, the parameters of the second policy network of the task processing model are updated to obtain the third update result; Based on the second update result, the parameters of the second evaluation network of the task processing model are updated to obtain the fourth update result.

4. The task processing method according to claim 3, characterized in that, Based on the first update result, the parameters of the second policy network of the task processing model are updated to obtain a third update result, including: according to The parameters of the second policy network of the task processing model are updated to obtain the third update result; in, The parameters of the second policy network for the task processing model are... This is the first updated result. These are preset variables.

5. The task processing method according to claim 3, characterized in that, Based on the second update result, the parameters of the second evaluation network of the task processing model are updated to obtain the fourth update result, including: according to The parameters of the second evaluation network of the task processing model are updated to obtain the fourth update result; in, The parameters are for the second evaluation network of the task processing model. This is the second update result. These are preset variables.

6. A task processing device, characterized in that, The device includes: The acquisition module is used to acquire state space data of at least one terminal interacting with at least one edge computing device during a preset time series; the edge computing device includes a base station and an edge server located next to the base station; the task includes an application generated by terminal i in time slot t. Multiple independent subtasks ; The processing module is used to input the state space data at the first moment in the preset time series into the first processing network of the task processing model for processing to obtain the first evaluation data. The state space data at the second moment in the preset time series is input into the second processing network of the task processing model for processing to obtain the second evaluation data; the second moment is greater than the first moment. Based on the first evaluation data and the second evaluation data, the parameters of the task processing model are updated to obtain an update result; wherein, the first processing network includes a first policy network and a first evaluation network; Specifically, the state space data at the first moment in the preset time series is input into the first processing network of the task processing model for processing to obtain the first evaluation data, including: The state space data at the first moment is input into the first policy network for processing to obtain the action state data at the first moment; The state space data and action state data at the first moment are input into the first evaluation network for processing to obtain the first evaluation data; The process includes, after inputting the state space data at the first moment into the first policy network for processing to obtain the action state data at the first moment, the method further includes: Based on the action status data at the first moment, determine the latency and energy consumption data when at least one terminal interacts with at least one edge computing device. When terminals and edge computing devices interact with each other, determine the impact of local computing queues, local transmission queues, edge computing queues, and edge transmission queues on task processing; The process of determining the local computing queue includes: The latency required for locally computed subtasks is determined by the waiting latency in the queue. Service delay composition: ； ； in, To calculate the time required for queue processing locally, This represents the number of time slots the subtask needs to wait in the local queuing system. ξ represents the time when all subtasks placed in the local computation queue before time slot t have been processed; ξ is the central processing unit (CPU) processing density, representing the number of clock cycles required for each bit of the processor to execute. express Input data size, This represents the amount of load newly arriving in the application at the start of time slot t. This represents the number of CPU revolutions per unit of time. for The actual selected uninstallation rate; For computation queues or transmission queues, if the processing or transmission of a subtask is completed within a time slot, then the next subtask in the queue will be processed or transmitted at the beginning of the next time slot. equal Subtract the arrival time of the subtask and add 1. The following formula indicates when the subtask will be in the local computation queue. The internal organs have been processed. ； in, This represents the floor function. This indicates the number of time slots required for the subtask to be processed in the local computation queue, at time 0. Initialization occurs during which the subtasks do not need to wait, therefore Therefore ; The energy consumed by the terminal device when performing local computing tasks is: ;in, This represents the energy consumed per CPU cycle locally, with a value of [value missing]. ; The process of determining the local transmission queue includes: The time required to send data from the local transmission queue to the edge server's computation or transmission queue consists of the local transmission queue's waiting latency and the transmission latency: ； ； in, express The number of time slots that a subtask destined for the edge server needs to wait in the local transmission queue. This indicates the time slot in which all subtasks placed in the transmission queue before time slot t have been processed; Δ is the time slot length. Let be the transmission rate between user i and base station m in time slot t; The subtask will be in the local transfer queue The internal organs have been processed. ； in, This indicates the number of time slots required for a subtask to be processed in the local transmission queue, at time 0. Initialization, at this time Therefore ; The energy consumed by the terminal device when transmitting tasks locally is: ;in, Indicates the terminal transmission power; The process of determining the edge computing queue includes: Assume the time when the unloaded subtask arrives at the edge computing queue is The time required for a task to be processed in the edge server's computing queue consists of the task processing time and the waiting time in the computing queue, i.e.: ； in, This represents the number of time slots that the unloaded subtask needs to wait in the edge computing queue, and it is calculated as follows: ; in, Indicates in time slot The initial length of the queue is updated using the following formula: ; in, Is edge server m in time slot Initial queue length, express The sum of tasks from all terminal devices arriving at MEC server m at any given time. This indicates the amount of data that edge server m can process in a single time slot; Define a variable To represent devices in the edge computing queue and equipment The processing order of the two tasks; ； Therefore Indicates all and The subtasks arrive at the same edge server computation queue simultaneously and are computed in a priority order. The sum of the quantities of the other subtasks of the subtask; The energy consumed by the terminal device when the edge server is computing tasks is: ； The process of determining the edge transmission queue includes: In the time slot t At the beginning, when the new application Upon arrival, it is either processed locally or offloaded to an attached edge computing device, or further forwarded to other nearby edge computing devices via a zero-latency wired link in the network; Tasks that need to be transferred to nearby edge computing devices are placed in the transmission queue of the edge server configured on the directly connectable edge computing device. Assuming the time when the unloaded subtask arrives at the edge transmission queue is... ; Edge computing devices transmit information via wired connections at a transmission rate of [missing information]. The time required for the uninstalled application to be forwarded from the currently directly connectable edge computing device to the target edge server is expressed as: ； ； in, This indicates the number of time slots the unloaded subtask needs to wait in the edge server's transmission queue. Indicates in time slot The time slot in which all subtasks previously placed in the MEC server's transmission queue have been processed; the subtasks will then be in the edge server's transmission queue. The internal organs have been processed. ; in, This indicates the number of time slots required for the unloaded subtask to be processed in the edge server's transmission queue, at time 0. Initialization is performed at this time. Therefore ; The energy consumed by the terminal device when transmitting tasks to the edge server is: ;in, This indicates the power of the device when it is idle.

7. A computing device, characterized in that, include: A processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method as described in any one of claims 1 to 5.

8. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 5.

Images